src/cpu/inorder/resources/fetch_seq_unit.cc

   1 /*
   2  * Copyright (c) 2007 MIPS Technologies, Inc.
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions are
   7  * met: redistributions of source code must retain the above copyright
   8  * notice, this list of conditions and the following disclaimer;
   9  * redistributions in binary form must reproduce the above copyright
  10  * notice, this list of conditions and the following disclaimer in the
  11  * documentation and/or other materials provided with the distribution;
  12  * neither the name of the copyright holders nor the names of its
  13  * contributors may be used to endorse or promote products derived from
  14  * this software without specific prior written permission.
  15  *
  16  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  17  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  18  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  19  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  20  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  22  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  26  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  27  *
  28  * Authors: Korey Sewell
  29  *
  30  */
  31
  32 #include "config/the_isa.hh"
  33 #include "cpu/inorder/resources/fetch_seq_unit.hh"
  34 #include "cpu/inorder/resource_pool.hh"
  35
  36 using namespace std;
  37 using namespace TheISA;
  38 using namespace ThePipeline;
  39
  40 FetchSeqUnit::FetchSeqUnit(std::string res_name, int res_id, int res_width,
  41                        int res_latency, InOrderCPU *_cpu, ThePipeline::Params *params)
  42     : Resource(res_name, res_id, res_width, res_latency, _cpu),
  43       instSize(sizeof(MachInst))
  44 {
  45     for (ThreadID tid = 0; tid < ThePipeline::MaxThreads; tid++) {
  46         delaySlotInfo[tid].numInsts = 0;
  47         delaySlotInfo[tid].targetReady = false;
  48
  49         pcValid[tid] = false;
  50         pcBlockStage[tid] = 0;
  51
  52         squashSeqNum[tid] = (InstSeqNum)-1;
  53         lastSquashCycle[tid] = 0;
  54     }
  55 }
  56
  57 FetchSeqUnit::~FetchSeqUnit()
  58 {
  59     delete [] resourceEvent;
  60 }
  61
  62 void
  63 FetchSeqUnit::init()
  64 {
  65     resourceEvent = new FetchSeqEvent[width];
  66
  67     initSlots();
  68 }
  69
  70 void
  71 FetchSeqUnit::execute(int slot_num)
  72 {
  73     // After this is working, change this to a reinterpret cast
  74     // for performance considerations
  75     ResourceRequest* fs_req = reqMap[slot_num];
  76     DynInstPtr inst = fs_req->inst;
  77     ThreadID tid = inst->readTid();
  78     int stage_num = fs_req->getStageNum();
  79     int seq_num = inst->seqNum;
  80
  81     fs_req->fault = NoFault;
  82
  83     switch (fs_req->cmd)
  84     {
  85       case AssignNextPC:
  86         {
  87             if (pcValid[tid]) {
  88
  89                 if (delaySlotInfo[tid].targetReady &&
  90                     delaySlotInfo[tid].numInsts == 0) {
  91                     // Set PC to target
  92                     PC[tid] = delaySlotInfo[tid].targetAddr; //next_PC
  93                     nextPC[tid] = PC[tid] + instSize;        //next_NPC
  94                     nextNPC[tid] = PC[tid] + (2 * instSize);
  95
  96                     delaySlotInfo[tid].targetReady = false;
  97
  98                     DPRINTF(InOrderFetchSeq, "[tid:%i]: Setting PC to delay slot target\n",tid);
  99                 }
 100
 101                 inst->setPC(PC[tid]);
 102                 inst->setNextPC(PC[tid] + instSize);
 103                 inst->setNextNPC(PC[tid] + (instSize * 2));
 104
 105 #if ISA_HAS_DELAY_SLOT
 106                 inst->setPredTarg(inst->readNextNPC());
 107 #else
 108                 inst->setPredTarg(inst->readNextPC());
 109 #endif
 110                 inst->setMemAddr(PC[tid]);
 111                 inst->setSeqNum(cpu->getAndIncrementInstSeq(tid));
 112
 113                 DPRINTF(InOrderFetchSeq, "[tid:%i]: Assigning [sn:%i] to PC %08p, NPC %08p, NNPC %08p\n", tid,
 114                         inst->seqNum, inst->readPC(), inst->readNextPC(), inst->readNextNPC());
 115
 116                 if (delaySlotInfo[tid].numInsts > 0) {
 117                     --delaySlotInfo[tid].numInsts;
 118
 119                     // It's OK to set PC to target of branch
 120                     if (delaySlotInfo[tid].numInsts == 0) {
 121                         delaySlotInfo[tid].targetReady = true;
 122                     }
 123
 124                     DPRINTF(InOrderFetchSeq, "[tid:%i]: %i delay slot inst(s) left to"
 125                             " process.\n", tid, delaySlotInfo[tid].numInsts);
 126                 }
 127
 128                 PC[tid] = nextPC[tid];
 129                 nextPC[tid] = nextNPC[tid];
 130                 nextNPC[tid] += instSize;
 131
 132                 fs_req->done();
 133             } else {
 134                 DPRINTF(InOrderStall, "STALL: [tid:%i]: NPC not valid\n", tid);
 135                 fs_req->setCompleted(false);
 136             }
 137         }
 138         break;
 139
 140       case UpdateTargetPC:
 141         {
 142             if (inst->isControl()) {
 143                 // If it's a return, then we must wait for resolved address.
 144                 if (inst->isReturn() && !inst->predTaken()) {
 145                     cpu->pipelineStage[stage_num]->toPrevStages->stageBlock[stage_num][tid] = true;
 146                     pcValid[tid] = false;
 147                     pcBlockStage[tid] = stage_num;
 148                 } else if (inst->isCondDelaySlot() && !inst->predTaken()) {
 149                 // Not-Taken AND Conditional Control
 150                     DPRINTF(InOrderFetchSeq, "[tid:%i]: [sn:%i]: [PC:%08p] Predicted Not-Taken Cond. "
 151                             "Delay inst. Skipping delay slot and  Updating PC to %08p\n",
 152                             tid, inst->seqNum, inst->readPC(), inst->readPredTarg());
 153
 154                     DPRINTF(InOrderFetchSeq, "[tid:%i] Setting up squash to start from stage %i, after [sn:%i].\n",
 155                             tid, stage_num, seq_num);
 156
 157                     inst->bdelaySeqNum = seq_num;
 158                     inst->squashingStage = stage_num;
 159
 160                     squashAfterInst(inst, stage_num, tid);
 161                 } else if (!inst->isCondDelaySlot() && !inst->predTaken()) {
 162                     // Not-Taken Control
 163                     DPRINTF(InOrderFetchSeq, "[tid:%i]: [sn:%i]: Predicted Not-Taken Control "
 164                             "inst. updating PC to %08p\n", tid, inst->seqNum,
 165                             inst->readNextPC());
 166 #if ISA_HAS_DELAY_SLOT
 167                     ++delaySlotInfo[tid].numInsts;
 168                     delaySlotInfo[tid].targetReady = false;
 169                     delaySlotInfo[tid].targetAddr = inst->readNextNPC();
 170 #else
 171                     assert(delaySlotInfo[tid].numInsts == 0);
 172 #endif
 173                 } else if (inst->predTaken()) {
 174                     // Taken Control
 175 #if ISA_HAS_DELAY_SLOT
 176                     ++delaySlotInfo[tid].numInsts;
 177                     delaySlotInfo[tid].targetReady = false;
 178                     delaySlotInfo[tid].targetAddr = inst->readPredTarg();
 179
 180                     DPRINTF(InOrderFetchSeq, "[tid:%i]: [sn:%i] Updating delay slot target "
 181                             "to PC %08p\n", tid, inst->seqNum, inst->readPredTarg());
 182                     inst->bdelaySeqNum = seq_num + 1;
 183 #else
 184                     inst->bdelaySeqNum = seq_num;
 185                     assert(delaySlotInfo[tid].numInsts == 0);
 186 #endif
 187
 188                     inst->squashingStage = stage_num;
 189
 190                     DPRINTF(InOrderFetchSeq, "[tid:%i] Setting up squash to start from stage %i, after [sn:%i].\n",
 191                             tid, stage_num, inst->bdelaySeqNum);
 192
 193                     // Do Squashing
 194                     squashAfterInst(inst, stage_num, tid);
 195                 }
 196             } else {
 197                 DPRINTF(InOrderFetchSeq, "[tid:%i]: [sn:%i]: Ignoring branch target update "
 198                         "since then is not a control instruction.\n", tid, inst->seqNum);
 199             }
 200
 201             fs_req->done();
 202         }
 203         break;
 204
 205       default:
 206         fatal("Unrecognized command to %s", resName);
 207     }
 208 }
 209
 210 inline void
 211 FetchSeqUnit::squashAfterInst(DynInstPtr inst, int stage_num, ThreadID tid)
 212 {
 213     // Squash In Pipeline Stage
 214     cpu->pipelineStage[stage_num]->squashDueToBranch(inst, tid);
 215
 216     // Squash inside current resource, so if there needs to be fetching on same cycle
 217     // the fetch information will be correct.
 218     // squash(inst, stage_num, inst->bdelaySeqNum, tid);
 219
 220     // Schedule Squash Through-out Resource Pool
 221     cpu->resPool->scheduleEvent((InOrderCPU::CPUEventType)ResourcePool::SquashAll, inst, 0);
 222 }
 223 void
 224 FetchSeqUnit::squash(DynInstPtr inst, int squash_stage,
 225                      InstSeqNum squash_seq_num, ThreadID tid)
 226 {
 227     DPRINTF(InOrderFetchSeq, "[tid:%i]: Updating due to squash from stage %i.\n",
 228             tid, squash_stage);
 229
 230     InstSeqNum done_seq_num = inst->bdelaySeqNum;
 231     Addr new_PC = inst->readPredTarg();
 232
 233     if (squashSeqNum[tid] <= done_seq_num &&
 234         lastSquashCycle[tid] == curTick) {
 235         DPRINTF(InOrderFetchSeq, "[tid:%i]: Ignoring squash from stage %i, since"
 236                 "there is an outstanding squash that is older.\n",
 237                 tid, squash_stage);
 238     } else {
 239         squashSeqNum[tid] = done_seq_num;
 240         lastSquashCycle[tid] = curTick;
 241
 242         // If The very next instruction number is the done seq. num,
 243         // then we haven't seen the delay slot yet ... if it isn't
 244         // the last done_seq_num then this is the delay slot inst.
 245         if (cpu->nextInstSeqNum(tid) != done_seq_num &&
 246             !inst->procDelaySlotOnMispred) {
 247             delaySlotInfo[tid].numInsts = 0;
 248             delaySlotInfo[tid].targetReady = false;
 249
 250             // Reset PC
 251             PC[tid] = new_PC;
 252             nextPC[tid] = new_PC + instSize;
 253             nextNPC[tid] = new_PC + (2 * instSize);
 254
 255             DPRINTF(InOrderFetchSeq, "[tid:%i]: Setting PC to %08p.\n",
 256                     tid, PC[tid]);
 257         } else {
 258 #if !ISA_HAS_DELAY_SLOT
 259             assert(0);
 260 #endif
 261
 262             delaySlotInfo[tid].numInsts = 1;
 263             delaySlotInfo[tid].targetReady = false;
 264             delaySlotInfo[tid].targetAddr = (inst->procDelaySlotOnMispred) ? inst->branchTarget() : new_PC;
 265
 266             // Reset PC to Delay Slot Instruction
 267             if (inst->procDelaySlotOnMispred) {
 268                 PC[tid] = new_PC;
 269                 nextPC[tid] = new_PC + instSize;
 270                 nextNPC[tid] = new_PC + (2 * instSize);
 271             }
 272
 273         }
 274
 275         // Unblock Any Stages Waiting for this information to be updated ...
 276         if (!pcValid[tid]) {
 277             cpu->pipelineStage[pcBlockStage[tid]]->toPrevStages->stageUnblock[pcBlockStage[tid]][tid] = true;
 278         }
 279
 280         pcValid[tid] = true;
 281     }
 282
 283     Resource::squash(inst, squash_stage, squash_seq_num, tid);
 284 }
 285
 286 FetchSeqUnit::FetchSeqEvent::FetchSeqEvent()
 287     : ResourceEvent()
 288 { }
 289
 290 void
 291 FetchSeqUnit::FetchSeqEvent::process()
 292 {
 293     FetchSeqUnit* fs_res = dynamic_cast<FetchSeqUnit*>(resource);
 294     assert(fs_res);
 295
 296     for (int i=0; i < MaxThreads; i++) {
 297         fs_res->PC[i] = fs_res->cpu->readPC(i);
 298         fs_res->nextPC[i] = fs_res->cpu->readNextPC(i);
 299         fs_res->nextNPC[i] = fs_res->cpu->readNextNPC(i);
 300         DPRINTF(InOrderFetchSeq, "[tid:%i]: Setting PC:%08p NPC:%08p NNPC:%08p.\n",
 301                 fs_res->PC[i], fs_res->nextPC[i], fs_res->nextNPC[i]);
 302
 303         fs_res->pcValid[i] = true;
 304     }
 305
 306     //cpu->fetchPriorityList.push_back(tid);
 307 }
 308
 309
 310 void
 311 FetchSeqUnit::activateThread(ThreadID tid)
 312 {
 313     pcValid[tid] = true;
 314
 315     PC[tid] = cpu->readPC(tid);
 316     nextPC[tid] = cpu->readNextPC(tid);
 317     nextNPC[tid] = cpu->readNextNPC(tid);
 318
 319     cpu->fetchPriorityList.push_back(tid);
 320
 321     DPRINTF(InOrderFetchSeq, "[tid:%i]: Reading PC:%08p NPC:%08p NNPC:%08p.\n",
 322             tid, PC[tid], nextPC[tid], nextNPC[tid]);
 323 }
 324
 325 void
 326 FetchSeqUnit::deactivateThread(ThreadID tid)
 327 {
 328     delaySlotInfo[tid].numInsts = 0;
 329     delaySlotInfo[tid].targetReady = false;
 330
 331     pcValid[tid] = false;
 332     pcBlockStage[tid] = 0;
 333
 334     squashSeqNum[tid] = (InstSeqNum)-1;
 335     lastSquashCycle[tid] = 0;
 336
 337     list<ThreadID>::iterator thread_it = find(cpu->fetchPriorityList.begin(),
 338                                               cpu->fetchPriorityList.end(),
 339                                               tid);
 340
 341     if (thread_it != cpu->fetchPriorityList.end())
 342         cpu->fetchPriorityList.erase(thread_it);
 343 }
 344
 345 void
 346 FetchSeqUnit::suspendThread(ThreadID tid)
 347 {
 348     deactivateThread(tid);
 349 }
 350
 351 void
 352 FetchSeqUnit::updateAfterContextSwitch(DynInstPtr inst, ThreadID tid)
 353 {
 354     pcValid[tid] = true;
 355
 356     if (cpu->thread[tid]->lastGradIsBranch) {
 357         /** This function assumes that the instruction causing the context
 358          *  switch was right after the branch. Thus, if it's not, then
 359          *  we are updating incorrectly here
 360          */
 361         assert(cpu->thread[tid]->lastBranchNextPC == inst->readPC());
 362
 363         PC[tid] = cpu->thread[tid]->lastBranchNextNPC;
 364         nextPC[tid] = PC[tid] + instSize;
 365         nextNPC[tid] = nextPC[tid] + instSize;
 366     } else {
 367         PC[tid] = inst->readNextPC();
 368         nextPC[tid] = inst->readNextNPC();
 369         nextNPC[tid] = inst->readNextNPC() + instSize;
 370     }
 371
 372     DPRINTF(InOrderFetchSeq, "[tid:%i]: Updating PCs due to Context Switch."
 373             "Assigning  PC:%08p NPC:%08p NNPC:%08p.\n", tid, PC[tid],
 374             nextPC[tid], nextNPC[tid]);
 375 }